Today’s Goals

1. Review 2023 Issues

• What went well, what didn’t

2. Identify Priorities for 2026

• Must address vs nice-to-have:

  • Spatial: 6R vs 9R?

  • Tag mixing: 1Q/2Q/3Q?

  • Some uncertainty around input data

Stock is Healthy

• SB/SBF=0 = 0.35 (0.30-0.40)
  • Above limit reference point
• F/FMSY = 0.59 (0.46-0.74)
  • Below limit
• Status:
  • NOT overfished
  • NOT overfishing
• 54-model grid:
  • All models within the green zone

Catch-conditioned model:

• Reduced from 11,421 to 3,067 parameters
• More stable convergence

sdmTMB CPUE standardization:

• Spatial-temporal modeling framework
• Replaced VAST approach

Lorenzen natural mortality:

• Size/age-dependent M scaling
• More realistic mortality patterns

Growth estimation:

• Von Bertalanffy parameters internally estimated
• Use of conditional age-at-length data

Uncertainty grid expanded:

• Both structural & parameter uncertainties incorporated

Improved model convergence:

• Diagnostic model achieved positive definite Hessian
• 48 of 54 grid models converged

Key Challenges

• Model instability (sensitivity to starting values)
• Limited age data (1,004 otoliths)
• Spatial structure complexity (9 regions)
• Size composition conflicts with tagging
• Tag mixing period uncertain

SC19 Evaluation(WCPFC 2023):

• “More rigorous… but model instability remains”
• Result: 14 research recommendations

Data Collection & Quality

• #1 Continue age & tagging data collection
• #2 Develop additional CPUE index series
• #3 Account for effort creep
• #11 Review size composition representativeness
• #12 Investigate weight vs tagging data inversion

Model Structure

• #4 Simulation study: spatial structure (6-region)
• #9 Identify correlated/sensitive parameters
• #13 Reduce sensitivity to starting conditions

Biological Parameters

• #5 Investigate unfished SSB increase (tropics)
• #6 Yield-per-recruit analyses
• #7 Evaluate growth/mortality variability
• #10 Explore seasonal/regional growth

Diagnostics & Sensitivity

• #8 Additional biological sensitivities
• #14 Tag mixing diagnostics

Problem

• Only 1,004 otoliths

• Limited spatial and temporal coverage (mostly 2009-2018)

Impact

• Maybe insufficient for internal growth estimation

SC19-#1, #7, #10

• Continue age data collection

• evaluate growth/mortality variability

• explore regional traits

Possible improvements

• Add more otoliths collected by Japanese colleagues

• Use tagging data to estimate growth?

Problem

• This complexity may have caused:
  • Convergence issues (parameter identifiability issues)

Evidence Supporting 6-Region (Hamer et al. 2023):

• Low tropical \(\leftrightarrow\) temperate movement

• High mixing within tropical/temperate zones

SC19-#4

• Simulation study: 6-region vs 9-region performance

• “areas-as-fleets” approach

Possible Approach

• Include 6R as a grid (or as one-off sensitivity)
  • Time constraints may hinder rigorous simulation testing given other priorities and complexity of the models

Problem

• Unknown mixing period (current: 2Q, more tests needed)
• Varies by location, size, season, etc.
• Most influential grid axis in 2023

SC19-#14

• Develop diagnostics; test 1Q/2Q/3Q; quantify uncertainty

Possible Approach

• Use estimates from IKAMOANA tag mixing analysis

Problem

• No iterative reweighting in 2023:
  • Fixed weights (no iterative weighting due to time constraints)
  • Data conflicts unresolved:
    • Weight comp vs length/tagging pull opposite directions

SC19-#11, #12

• Review representativeness; investigate inversion

Possible Approaches

• Explore Francis reweighting method and Dirichlet-multinomial approaches
• Convert weight comp to length comp externally

Problem

• Unfished SSB increases over time in tropical regions (3,4,7,8)

Potential Causes

• Regional recruitment allocation patterns
• Movement misspecification
• Initial population distribution assumptions

SC19-#5

Investigate model specifications causing this trend

Possible Approaches

• Change in spatial structure (6R vs 9R)
• Adding Effort creep in CPUE indices
• Penalize extreme recruitment deviations

Problems

• Sensitivity to initial conditions:
  • Models achieve suboptimal local minima
• Jittering to find better convergence:
  • Jittering normally used for diagnostics (robustness check)
  • Every grid model improved through jittering (model not at global minimum)

SC19-#9, #13:

• Identify correlated/sensitive parameters; strategies to decrease starting value sensitivity

Possible Approaches

• Model simplification
• Self-consistency checks to investigate which parameters are unidentifiable

Jittering in 2023 Assessment (Day et al. 2023)

All code for analysis and report publicly available:

• Analysis: github.com/PacificCommunity/ofp-sam-2026-BET
  • MFCL model runs
  • Plot generation
• Report: github.com/PacificCommunity/ofp-sam-2026-BET-report
  • Presentation slides (including this Pre-PAW)
  • SC report
  • Plots and tables in html format

Containerized Environment

Docker image for reproducibility:

• Pre-built R environment with all dependencies
• Ensures identical results across platforms
• Available at: ghcr.io/pacificcommunity/bet-2026

Automated Workflows

GitHub Actions for quality control and continuous integration:

• Automatic report compilation
  • Every update on writing triggers rebuild, including figures
  • Ensures report stays synchronized with analysis
• Code testing
  • Diagnostic model (with a single iteration in MFCL) runs on each commit
  • Plot generation tested
  • Reproducibility verified

Transparent Proces

• Community can review methods
• Results fully reproducible
• Facilitates review and collaboration
• Supports future assessments